what is the meaning of number in each model

#17
by martinKH - opened

when i download llama-3.1-8b. there is only one parameter file, one ckpt file, one tokenizer file

so, it is clear to set-up to generating task.

And i download llama-3.1-70b at few minute ago. there is multiple ckpt file. and there is no explanation about it. (i searched a lot)
ex) consolidated.00.pth, consolidated.01.pth ... consolidated.07.pth

but it seems like 70B model splited in multiple ckpt file.

If so, how can i make it in a one file with multiple ckpt file.

if there is code. someone help me

Sign up or log in to comment